Reimplementation / alternative to no_globals based on static analysis

[dlukes]

Code objects tell you which names are global and which are local:

def foo():
    bar = 1
    baz = 2
    print(bar, baz, qux)

print(foo.__code__.co_names)
print(foo.__code__.co_varnames)

('print', 'qux')
('bar', 'baz')

Ideally, this would be an IPython extension that you enable once and then it continuously warns you whenever you create a function that refers to (non-module, non-callable -- by default, but customizable) globals.

[dlukes]

There's also co_freevars for free variables, i.e. variables captured by a closure from an enclosing scope. (And conversely, enclosing functions keep track of what variables nested functions capture via co_cellvars, but that's not directly relevant here.)

[dlukes]

And a ton of additional useful exploration here: https://amakelov.github.io/blog/deps/

The blog post doesn't recommend static analysis, but the considerations based on which it decides against it don't apply here I think. Dependency tracking for caching purposes really needs to be 100% accurate, whereas the purpose here is just to warn about very basic cases of forgetting to delete a global variable use (or rename it).

[dlukes]

Possibly also warn about currently undefined global variables, which would result in NameError at runtime.

[dlukes]

Just incidentally -- to get the value of a free variable i captured by closure f:

f.__closure__[f.__code__.co_freevars.index("i")].cell_contents